VietTESOL International Convention 2026

Generative Artificial Intelligence (GenAI) is increasingly integrated into EFL classrooms to support digital multimodal composition (DMC). However, limited evidence explains why learners with similar proficiency levels experience GenAI-assisted tasks differently, particularly in affective engagement and vocabulary learning. This qualitative study investigates (1) how similarly proficient EFL learners differ in affective engagement during a GenAI-supported DMC pre-reading task and (2) how these differences relate to perceived vocabulary gains. Twenty Vietnamese undergraduates at CEFR B1 completed a DMC pre-reading task using ChatGPT to generate inquiry questions, clarify unfamiliar vocabulary, and create thematic illustrations, followed by a reading comprehension activity and an 11-item open-ended reflection. Data were analyzed thematically using Keller’s ARCS model (Attention, Relevance, Confidence, Satisfaction). Findings revealed substantial divergence despite identical task conditions. Most participants reported that vivid AI-generated visuals increased attention and helped topic comprehension, while some perceived outputs as generic and weakly relevant. Confidence was the most differentiating factor: students who iteratively refined prompts (typically two to three rounds) reported stronger self-efficacy and readiness for reading than one-shot prompters. Satisfaction and perceived vocabulary gains were highest when GenAI outputs aligned closely with reading content and learning goals, but decreased when outputs were unrealistic or cognitively overloaded. The study highlights prompt-iteration training and output-task alignment as key pedagogical conditions for effective GenAI-supported vocabulary learning.